Hydrol. Earth Syst. Sci.
, 13, 19391951, 2009
www.hydrol-earth-syst-sci.net/13/1939/2009/
Author(s) 2009. This work is distributed under
the Creative Commons Attribution 3.0 License.
HESS Opinions
Classification of hydrological models for flood management
E. J. Plate
Emeritus Professor, Hydrology and Water Resources Planning, Universitat Karlsruhe, Germany
Received: 2 June 2009 Published in Hydrol. Earth Syst. Sci. Discuss.: 1 July 2009
Revised: 22 September 2009 Accepted: 22 September 2009 Published: 20 October 2009
Abstract. Hydrological models for flood management are
components of flood risk management, which is the set of
actions to be taken to prevent flood disasters. It is a cyclical
process: initiated by occurrence of an extreme flood it leads
through the reconstruction and rehabilitation phase to risk
assessment and project planning and implementation, and fi-
nally to operation and preparedness for a next extreme flood
when the cycle starts again. We subdivide the tasks of flood
management into two consecutive parts: planning and oper-
ation, which basically require different kinds of hydrological
models. For planning, real time runoff is not needed, one
works with design scenarios. For this task models should be
used appropriate to the tasks at hand, which reflect character-
istics of landscape as well as of hydrological scale. For op-
eration, hydrological forecast models are needed which have
to meet a different set of conditions. In this paper, require-
ments for hydrological models as functions of application,
geology and topography, and of area size are surveyed and
classified, as a first approach for guiding users to the correct
type of model to be used in a given location. It is suggested
that one always should start flood modeling with an analy-
sis of local conditions and select or develop task and locality
specific models.
1 Hydrological tasks for flood risk management
Recent large floods in many regions of the world have cre-
ated new awareness for the need of systematic approaches
to flood disaster prevention. In response to this need flood
risk management has developed as a method, which system-
atically covers all actions for obtaining and managing fea-
sible and financially affordable protection measures against
Correspondence to: E. J. Plate
(plate@iwk.uka.de)
Published by Copernicus Publications on behalf of the European Geosciences Union.
Hydrology and
Earth System
Sciences
floods. It includes not only measures for protection of peo-
ple and goods at risk, but also for conservation of environ-
ment and riparian ecology. Modern design principles include
the requirement that non-technical measures, including mea-
sures of temporary protection should be used wherever pos-
sible. This approach has been promoted world wide by the
International Decade of Natural Disaster Reduction of the
United Nations (UN/ISDR, 2004).
Hydrological tools for these actions are flood forecast
models and models to determine design floods for flood pro-
tection measures. Prerequisite for many temporary flood pro-
tection measures is a good forecast of expected flood levels,
whereas design for permanent measures requires flood levels
for different exceedance probabilities. A survey of require-
ments for models for flood risk management is given in this
paper, which is intended as a first approach towards a sys-
tematic determination of the kind of model to be used for a
specific flood problem in a specific location, and not as a sur-
vey of existing models, for which excellent recent summaries
are available (see for example Singh and Woolhiser, 2000).
1.1 Flood protection and risk management
Risk management must be seen as a cycle, as shown in Fig. 1.
This figure reflects the fact that there are two parts to risk
management, as has been described in detail by Plate (2000).
The lower half cycle covers the planning phase and includes
planning, design and project implementation. The upper half
reflects the operational phase, including maintenance, pre-
paredness, and response and recovery after an extreme event.
Although planning and operation are conducted by different
actors, it is necessary that they are considered together as
part of comprehensive flood risk management for each flood
prone location. This is implemented in new German regula-
tions and codes, such as the Directive for the determination of
the design flood of the German state of Baden-Wurttemberg
(LFU, 2005). Assume the risk management cycle to start
1940 E. J. Plate: Classification of hydrological models for flood management

Risk management:
Operation

Disaster

Preparedness Response

Maintenance (new) demand

societal
objectives

technical + administrative
capabilities

Mitigation Risk assessment

Planning of
measures: Defining hazards
(-technical (maps)
Hazard analysis
non - technical)
Decision making
Risk mitigation: Vulnerability analysis
Determination of risk
Implementation
Planning

Fig. 1. The cycle of risk management.

Fig.1: The cycle of risk management

with a destructive flood event in the region under study or
nearby. After a phase of relief and reconstruction as imme-
diate response to a flood disaster, the flooding situation is
reassessed and frequently leads to demands for an improved
protection system. A planning phase is initiated, in which
options for meeting these demands are identified and their ef-
fects evaluated. In particular, for areas that experience floods
only infrequently, it is necessary to also develop potential
damage scenarios for floods larger than design floods, or for
situations of breaking of dikes or dams. Damage assess-
ment methods for dam breaks such as developed by Beltamio
de Almeida et al. (2000) should also be used for dikes, al-
though consequences of dike breaks usually are less severe
than those from breaking dams that impound large reservoirs.
For each option risks must be determined through the pro-
cess of risk assessment, which combines hazards i.e. mag-
nitude of flood levels and their probability of being exceeded
with vulnerabilities, i.e. potential damages for each object
at risk buildings, highways, dikes etc. Hazards are deter-
mined and expressed in hazard maps, which show areas of
inundation as functions of flood levels of given exceedance
probability. Then the risk as expected value of damages
in the flooded areas is calculated, just as is done for indi-
vidual buildings by the insurance industry. This approach
has recently been formalized within the European Commu-
nity through the European Union Flood Directive (EU-FD,
EU, 2007) which requires that in accordance with principles
laid down in the Water Framework Directive (EU-WFD, EU
(2000) which requires basin wide planning) risk maps are to
be prepared within a specified time frame.
With risk as important decision criterion the process of de-
cision making is initiated. The EU-FD requires that plans are
drawn up for improving protection where needed, for this
task also setting a time frame, i.e. the degree of demanded
protection is established, plans to meet these demands by
technical or non-technical means are prepared by experts,
discussed by affected people and administrative bodies, and
finally decided on by the owner in case of a private project
or by responsible political decision makers in case of a
project in the public domain. Then the existing system is
improved, or a new system developed. For the operational
phase, finished systems are turned over to the system man-
agers staff, who not only have to maintain the system, but
who also have to adequately respond to forewarnings: they
have to produce and interpret forecasts from a flood fore-
cast system (if it exists) and warn people at risk immediately
before the next extreme event. Then the management cycle
starts again.
Listed in the centre of the risk management cycle are so-
cietal conditions under which flood risk management has
to be performed. They reflect the value system of the so-
ciety at risk, but also available technology, and scientific

Hydrol. Earth Syst. Sci., 13, 19391951, 2009 www.hydrol-earth-syst-sci.net/13/1939/2009/

E. J. Plate: Classification of hydrological models for flood management
understanding of the flood environment conditions which
change with time due to changes in climate, but mostly
due to changes in land use and habitation (a recent study by
Schumann et al., 2001 shows that today in Germany there
are no observable changes in flood runoff, which can be at-
tributed unequivocally to climate change). Because of these
changes, flood risk management is a task to be reconsidered
by every generation. Todays objective of flood protection is
to provide, with due consideration to environmental require-
ments and ecological and legal constraints, a safety against
the T -year flood, where T is the recurrence interval, and to
be prepared for floods that exceed this level.
Decisions based on recurrence intervals reflect an intuitive
assessment of the impact of protection system failure. On
large rivers in Germany i.e. on Rhine, Weser and Elbe
T =200 is selected, whereas protection against storm surges
in low lying countries, as in the Netherlands, may be as high
as T =10 000 or more. Furthermore, flood risk management
must include plans for handling residual risks, i.e. risks for
the case of protection system failure, caused either by floods
larger than the design flood, or by failures due to technical
defects or human error. The residual risk usually is defined
vaguely as the risk that exists even though protective mea-
sures are in place.
An important problem in modern flood risk management
is to put the decision process for flood safety on a more ob-
jective base, by using a quantitative determination of residual
risk as expected damage of failure of the protection system.
This aspect is not covered widely in flood management prac-
tice. To a first approximation, residual risks may be calcu-
lated based on the assumption that every flood exceeding the
design flood will cause the same damage, so that risk is prod-
uct of exceedance probability PE and damage K, where K is
usually quantified in terms of monetary units, i.e. US $. More
detailed analyses also consider dependency of damage on de-
sign flood level (Merz and Gocht, 2003), in which case risk
is the expected value of damage. Residual risks are used as
components of cost benefit analyses, where avoided dam-
age due to a protection system is one of the benefits, whereas
monetary risks and construction costs, properly discounted
(Loucks and Van Beek, 2005) are parts of the costs. Ideally,
the solution that minimizes the cost benefit ratio should be
selected, but in practice, monetary risk is not the only type of
risk to be considered in flood protection planning.
Ecological damages, as well as social consequences are
also important, although there are neither tested indicators
for quantifying these risks, nor weights which express rel-
ative importance of these indicators in comparison to mon-
etary risks. Indicators and weights are expressions of the
social value system of a society, which ultimately trans-
lates into political actions. For setting priorities for such ac-
tions indices could be useful, which should be functions of
weighted indicators. Assignment of weights to indicators is
a political task, whereas derivation of indices is a scientific
challenge (Birkmann, 2006). As is evident from this discus-
www.hydrol-earth-syst-sci.net/13/1939/2009/
1941
sion, flood risk management is a process, which requires nu-
merous actions at different levels and by many different per-
sons. It is not really a scientific process, because the role of
science is to identify causes and consequences and develop
tools, not make decisions on values. Among the tools which
science can contribute are hydrologic and hydraulic models,
which shall be considered in the remainder of this paper.
1.2 Models for operation vs. models for planning
Models for flood protection should be application oriented.
For the planning phase one needs models for developing
flood inundation and flood risk maps, or models for calcu-
lating water levels or discharges for the design of flood pro-
tection measures. Furthermore, in preparation for the opera-
tional phase, models are needed to determine operation rules,
for example for operation of reservoirs. Most reservoir oper-
ation rules are based on scenario calculations with historical
floods. However, today system operators want dynamic op-
erational models that can be used in real time for deciding
releases in anticipation of future floods, or for controlling
series of barrages for effective dynamic storage of flood wa-
ters, as needed for example on the upper Rhine, in order to
meet the protection target of the Integrated Rhine Program.
Flood forecasting models have to be developed, tested on his-
torical events, and put into service in the planning and imple-
mentation phase. Such models are also needed for decisions
on setting up temporary protection walls, or for evacuating
endangered population groups. Development of all plans
necessary for response to cases of extreme floods, which ex-
ceed the capacity of the protection system, are part of the
planning and implementation phase of flood risk manage-
ment.
Flood forecasting occurs in both phases of the flood risk
management cycle: during planning, the forecast model is
designed and calibrated, and during operation its successful
operation is prerequisite for any effective early warnings. Be-
cause an effective flood forecast and early warning system is
generally less expensive than technical measures, it often is
the most cost effective type of flood protection system, in
some cases the only one, in particular for many developing
countries. This is the case for some of the large rivers of Asia
or Africa, where floods are frequent and lead to large losses
of lives such as in 1998 on the Limpopo in Mozambique,
or in 2000 on the Mekong, where more than 3000 people
drowned. Flood forecasting is the chosen method for pre-
venting, or at least reducing, such losses of lives in the future.
It follows from these descriptions that there are two im-
portant categories of models to be used in flood risk man-
agement: forecast models and planning models. These two
types of models shall be discussed in the following sections.
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
1942 E. J. Plate: Classification of hydrological models for flood management
1.3 Forecast and prediction
The difference of planning vs. forecast models is illustrated
in Fig. 2 (from Plate, 2007). Objective is to forecast wa-
ter levels ha (t0 +TF ) at time TF later than the present time
t0 , where TF is the forecasting time and ha (t) is the actual
value of the water level at time t. A forecast model is used
to forecast a value hF (t0 +TF ). The forecast model must be
a function of the initial value ha (t0 ) at time t0 , at which the
forecast is made. Regardless of the forecast model used any
forecast is only an estimate, and for every forecast an error
band exists, which can be expressed by means of a pdf (prob-
ability density function) fh0 (t0 +TF ), which depends both on
h0 (t0 ) and on TF . The larger TF , the broader the error band
becomes, up to a limit when forecast times exceed a certain
maximum value TP , when the initial conditions become ir-
relevant, and forecasts degenerate into predictions (in the hy-
drological sense), i.e. hF (t0 +TF ) is a random variable with
f (h) independent of time and initial value.
For flood forecasting it is of major importance that the
forecast value is accurate. In many cases an erroneous fore-
cast is worse than no forecast at all. People who had trusted
a forecast that went wrong for example, that forced them
to evacuate an area will not likely trust a future forecast.
Consequently, development of dynamic models for real time
forecasting with as narrow an error band as possible is a ma-
jor challenge for hydrological research. At this time, the
output of most models is deterministic. An assessment of
the error for such models is usually done (if at all) through
sensitivity analyses or scenario development, in which the
range of possible values of the parameters of a model or of
the model inputs are estimated and the results analyzed. Tra-
ditional is the assessment of upper and lower bounds, but a
modern trend is to determine ensembles from many combi-
nations of probabilistically distributed parameters to obtain
estimates in terms of probability distributions of outputs of
the model, which then can be further analyzed to yield the en-
semble average and error bounds expressed in terms of stan-
dard deviations of the ensemble, i.e. Krzysztofowicz (2001).
Ensemble weather forecasts have a long tradition in meteo-
rology, however, the accuracy of meteorological forecasts of
rainfall is still the weakest link in improving flood forecast
models (Todini, 2004).
For designing technical flood protection systems only
good predictions of possible future extreme water levels for
given exceedance probabilities are needed, and time of oc-
currence is irrelevant. The classical approach is to use statis-
tical extreme value analysis of data obtained at river gages.
For basin wide measures this is not sufficient. Rainfall-runoff
models (RR-models) must supplement traditional extreme
value models for flood risk management. Hydrologists are
challenged to provide these models.
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
actual value of Q(t )
Q(t) uncertainty band for t 0
t crit
pdf of prediction of Q
h
2
h0 forecasted
value of Q(t 0 + TF )
forecast
time TF
t
prediction time TP at t 0
t
t0
Fig.2: Defining
Fig. 2. Defining forecasts forecasts and predictions
and predictions.
2 Rainfall-runoff models for flood management
Two types of RR-models for determination of floods of given
frequencies can be distinguished. One type uses rainfall
runoff modeling of the continuum of runoff in a river. Histor-
ical time series of rainfall (suitably area averaged) are used
and the resulting calculated runoff time series is compared
with the observed runoff time series. Differences between
values from observed time series and from RR-model can be
interpreted as realizations of a random process. Their mean
value is a measure of model bias to be corrected by pa-
rameter adjustment and their variance is a measure of un-
certainty. Because different sets of parameters may yield the
same variance, (a property called equifinality by Beven and
Freer, 2001), this method may yield good results on the av-
erage for the observed time series, but it may fail when ex-
trapolated, as is observed when the probability distribution of
extreme values of the observed time series of runoff at some
gage is compared with a distribution of extreme values of the
calculated series.
The second type of RR-model is event based. It is not in-
tended to be used for the whole time series. Its exclusive
purpose is to predict extreme values of runoff i.e. peaks,
18
volumes, and shapes of flood waves. When used for planning
purposes in flood risk management these models use hypo-
thetical rainfall fields. These are T -year area averaged rain-
fall fields that are more or less uniformly distributed over the
basin, under the assumption that the T -year area-averaged
rainfall will also cause the T -year flood. For practical ap-
plications of this method, the writer (see Plate et al., 1988)
has always insisted on validating such models at available
gauging stations against extreme value distributions of local
runoff.
All RR-models have in common that they have to describe
the physical transformation of rainfall into runoff. This re-
quires a common structure for all RR-models. This common
model structure will be discussed in the next section.
www.hydrol-earth-syst-sci.net/13/1939/2009/
E. J. Plate: Classification of hydrological models for flood management
2.1 Components of RR-models
Hydrological RR-models have three levels, each associated
with three different time scales as indicated in Fig. 3 (adapted
from Plate and Zehe, 2006). The data level consists of per-
manent, seasonal, and event based data. Permanent are ge-
ometric and geological properties of the basin: basin area,
topography and geology, river network and soil composition,
as well as properties that change only gradually, such as land
use: i.e. forest cover, road networks, urban developments,
or large scale climate. Parameters describing agricultural
activities and seasonal climate are associated with seasonal
time scales and determine seasonal variability of runoff co-
efficients. And finally, there are data for the event time scale.
These depend on the model purpose. Planning requires his-
torical data, whereas forecasting requires short term and real
time event data, with time increments ranging from minutes
to hours, depending on the size of the basin area.
The second level in Fig. 3 is the level of model formula-
tion and development. The process of model building should
start with a thorough assessment of the basin under inves-
tigation, orienting the model building process on available
hydro-meteorological data, but also on characteristics of the
basin, as well as on the basic requirements for decision mak-
ing as is expressed through the 4-th level of Fig. 3. The deci-
sion level determines the information needed, and level 3 is
thus determined by level 4.
Hydrologists charged with developing a flood planning or
forecast model for a basin should explore and describe its ge-
ological characteristics, trace its river networks, identify sur-
face and groundwater interactions. Important flow paths of
surface and subsurface flows need to be identified from the
beginning, and appropriately reproduced in the model. No
universal model exists that fits everywhere. For each situa-
tion and each catchment models should be built or adapted
appropriate to location and application. They should reflect
local conditions and incorporate all important human activi-
ties which may modify the rainfall runoff process. Due con-
sideration should also be given to different time scales of
different processes. Seasonal processes such as interactions
of groundwater and surface water may well be described by
models using larger time steps than runoff.
2.2 Types of rainfall-runoff models for flood
calculations
In view of what has been said in Sect. 2.1 the choice of a RR-
model is determined by intended application, basin scale, and
available data, which set conditions for development of a new
or adaptation of an existing model. The plethora of available
RR-models can be divided into three types: models based
on rectangular grids, models based on sub-catchments, and
models based on response units. Topography and geometry
of grid based models are derived from available large scale
digital terrain models (DTMs). From the elevation of the four
www.hydrol-earth-syst-sci.net/13/1939/2009/
1943
corners of a grid cell the slope of the cell is determined, and
the channel network is derived from these slopes by means of
special algorithms (Garbrecht and Martz, 1997; or Crowley
and He, 2005). Climate and land use variables are combined
with grid models through geographic information systems
(GIS). Catchment based models (CBMs), on the other hand
are vector oriented. They require subdividing the basin into
sub-catchments, whose sizes and topographic characteristics
have to be derived from DTMs. This is a lengthy prepara-
tory process, which has the advantage that transfer of land-
scape features from topographic maps is easier facilitated,
and river networks, geology, and land use can be naturally as-
sociated with basin features. The third type, models based on
response units (REM) divide catchments into units of equal
runoff formation. Both CBMs and REMs require that sub-
units are connected by means of networks of channels, and
all models require that due consideration be given to the hy-
draulics of the channels. Usually, not the discharge is needed
for flood studies, but the water level, which only for special
conditions can be inferred from stage discharge curves. In
most cases, it is necessary to convert discharges into stages
by means of hydraulic models, which range from stationary
1-D models to instationary 2-D models, incorporating flood
development over flood plains.
Grid based models are preferred for large scale continuous
models, such as for climate investigations, but they are also
applied frequently for flood modeling, both for flood fore-
casting (i.e. Todini, 1996) and for planning flood protection
systems. With their help the continuum of floods is deter-
mined from long term rainfall time series of observed and
area-averaged rainfall and water balances, including calcula-
tions of the time series of evapo-transpiration. Such models
are also useful for event based models, for design or for fore-
cast, in order to determine the initial moisture state of the
area element.
All area models have in common that they use a vertical
component for determination of that part of the storm rain-
fall which becomes flood runoff, and a horizontal component
for the routing of the rainfall excess to the nearest channel of
the river network. Models for runoff use runoff coefficients
ranging from simple constants which are empirically corre-
lated to soil and groundcover parameters, to sophisticated
functions obtained from water balance models, for which the
area element is represented by an equivalent vertical soil col-
umn consisting of different layers (i.e. Todini, 1996; Refs-
gaard and Storm, 1995; De Roo et al., 2000). Water balance
models separate the rainfall (minus interception) into surface
storage, and groundwater replenishment by means of an in-
filtration soil water transport model of varying complexity
(Liang et al., 1996; Todini, 1996; Crowley and He, 2005).
Runoff is routed from the area elements cells, sub-
catchments or REMs to the point of interest on the chan-
nel network. Routing models should reflect the considera-
tions of relative size of area element to channel network to
be discussed in Sect. 2.3. Simple models operate by using
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
1944 E. J. Plate: Classification of hydrological models for flood management

Data level
Geo-Information-Systems (GIS), Databanks GIS and Data banks Data files

Base variablsn Derived Distribution of

Base variables Derived Variables precipitation
Grid points, Channel- Climate variables
Topography, in space and
Plant cover
Soilproperties networks, Darcy coeff. ET(t) time, runoff
Geology Transmissivity Seasonaldistribution
of water quality root depth erosivity
Land use Erodibility etc. yield soil moisture

permanent scale seasonal scale event-

scale

Model level

Hydrological basic model Seasonal Transport model

Topography, Digital terrain models transport model Event based rainfall and
channel network, subcatchments, Seasonal distribution of runoff distribution.
long time balances of water and pollutants, herbicides, Distribution of pollutants
matter fertilizers and water and other agents

Output level
Tables and maps Output models
direct output
Decision level
Decision processes

Fig. 3. Levels of hydrological models

Fig.3:(adapted
Levels from Plate and Zehe,
of hydrological 2006).(adapted from PLATE & ZEHE, 2006)
models

only translation, assuming a constant velocity of runoff from
the element. More complex models are based on linear sys-
tems, applied to each element. For example, Crowley and
He (2005) use three parallel linear reservoirs, one for surface
runoff, one for interflow, and one for baseflow, or groundwa-
ter runoff.
For CBMs the RR process for each sub-catchment is de-
scribed by area models, which not only reflect the soil mois-
ture balance (i.e. Bronstert, 2005) but also incorporate dis-
tinctive catchment features, such as local topography and
land use such as urbanization and the network of roads and
railways. The connectedness of the sub-catchments follows
the channel network, in which runoff from sub-catchments is
routed downstream. Such a model can be very detailed, de-
pending on the resolution into sub-catchments, and the sub-
models selected for hydrological processes. How much de-
tail is to be incorporated will depend on the model purpose,
and on the scale of the region. Obviously, a flood model for
a basin of, say, 1000 km2 does not need the same resolution
as an area of a few hectares.
2.3 RR-modeling in different landscapes
Different characteristics of landscapes should require differ-
ent types of models. For example, floods in mountain val-
leys have very different characteristics from floods on flood
plains of large rivers. Theoretical hydrologists tend to use
the same type of model for all types of catchment, although
it seems obvious that the model should reflect the dominat-
ing processes for the type of landscape for which the model
is to be applied. We recommend to distinguish four differ-
ent types of landscapes and to develop models accordingly.
These are (a) high mountain ranges, (b) foothill ranges with
or without vegetation, (c) large flood plains, and (d) urban
areas, as indicated schematically in Fig. 4. A fifth region
is the area affected by coastal processes, for example delta
regions which are subjected to storm surges. Such a subdi-
vision is important for design of flood protection measures.
From a physical point of view, it is useful to further sub-
divide these area types. Within each of these areas there
exist sub-areas with their special hydrological characteris-
tics, for example forested regions, or wet lands etc. and
a subdivision of sub-catchments into such characteristic sub-
areas eventually leads to decomposition of sub-catchments
into many different REUs. Mountain areas are mainly threat-
ened by flash floods intensive and local rainfall events,
which lead to rapid increase of water levels and velocities in
runoff channels. In general, river courses in these areas are
deeply incised, and flooding usually is restricted to a narrow
strip along the river, where due to high velocities damages to
bank protection works and structures as in villages where
houses have been built too close to the creeks can19 be very
heavy, aggravated by frequent occurrence of debris jams, in
particular on bridges. Frequently extensive damage occurs

Hydrol. Earth Syst. Sci., 13, 19391951, 2009 www.hydrol-earth-syst-sci.net/13/1939/2009/

E. J. Plate: Classification of hydrological models for flood management
Mountain ranges
Foothill region
Dam
Bed
Plains rock
Urban
area
Aquifer
Fig. 4. Types of hydrological
Fig.4: Typesareas.
of hydrological areas
mainly on highways which for technical reasons had been
built along the rivers. Flood protection in such areas consists
at most of bank protection works, more usual is a flood pro-
tection strategy which on each side of the creek leaves a strip
of land where no human activity is permitted, or where land
use is restricted to agriculture. In such valleys, a detailed
analysis of floods is frequently of little use: planning models
in such areas usually are hydraulic models for extreme flood
scenarios based on historical floods.
In foothill regions, or in the geologically ancient moun-
tains which are typical for the State of Baden-Wuerttemberg,
in Germany, extreme precipitation or snow melt usually lead
to more widespread inundations than in mountain valleys,
and velocities are not of the same importance. Distinguish-
ing characteristics of floods in such regions is their impact on
villages and agricultural lands. Flood protection measures in
such areas consist of reduction of peak flows by means of
retention basins in the upper parts of the small rivers, and
removal of narrow sections in villages, with dikes in partic-
ularly sensitive stretches of the rivers or creeks. Flood pro-
tection of villages and small cities against the 10 to 30 year
flood is secured by these measures; floods of lower proba-
bility are reduced, but not avoided. Usually many retention
basins are used in a basin, each of which protects a small
area up to a level of T =20 to 50, and all of them in combina-
tion reduce flood frequencies in the lower parts of the rivers
to about T =100 years. Dikes, although standard measures
for flood protection for larger rivers, are rarely used on small
rivers (with catchments of a few hundred km2 ). For plan-
ning measures in foothill regions RR-models should be used,
whose structure depends on the catchments size, as will be
described in the next chapter.
In the plains of low lands velocities are even less impor-
tant. Damage is mainly caused by high water levels, and
in some situations due to interaction with groundwater
groundwater tables being raised in inundated areas, which in
www.hydrol-earth-syst-sci.net/13/1939/2009/
1945
turn flood basements or cause backing up of sewerage chan-
nels. The greatest threat to human lives comes from wide
spread inundations, in particular when over very wide areas
the water level rises only slowly, and escape routes are cut off
so that people are trapped on higher grounds, if help does not
come soon enough. Dikes are the natural measures for pro-
tecting low lands, but dikes may fail, or water levels reach
heights above the design height of the dike system. Today
one finds that in many parts of the world protective measures
include also widening of the river flood plain between dikes,
by new dike lines further inland, with a double dike system.
In some cases existing dikes are altogether removed, or the
formerly inundated flood plains are replaced by flood pold-
ers, which are flooded only when the water level in the main
river exceeds a certain critical value. Obviously, forecasting
future discharges or water levels for such areas is of consid-
erable importance not only for warning endangered popu-
lations, but also for the purpose of operating side polders or
retention basins in the catchment of the river.
Urban areas need special hydrological models, to incor-
porate sewer systems and runoff conditions from streets and
houses. Hydraulic RR-models are needed to describe flood-
ing from rainfall, as well as from rivers, on whose banks
cities are located. In Germany, smaller natural water courses
have been made part of the local sewer system, and many
of the small creeks flowing through cities or villages have
been confined into pipes. Extreme floods, often combined
with debris and trash plugging, cause such pipes to overflow
and produce heavy local flooding. Typical flood situations
were observed in the city of Dresden during the August flood
of 2002, which was caused both by flash floods from small
tributaries flowing into the city and bank overflow of river
Elbe. A third important cause of damage for Dresden was
the raised groundwater table due to inundation of the city.
Consequently, it is necessary that urban drainage models for
cities are integrated into detailed RR-models of rural areas,
not only to evaluate the effect of the basin river network on
urban flooding, but also to assess effects of urbanization on
runoff from catchments.
2.4 Hydrological scales and their significance in flood
calculations
The choice of a RR-model is not only dependent on type of
area, as described in the previous section. An additional deci-
sion criterion for model selection is the hydrological scale of
the area a subject which owes much of its scientific erudition
to Dooge (1986), but which will here be discussed from an
application oriented perspective. Hydrological scales have
become important
20 elements of categorizing basins (Plate,
1992; Bloschl and Sivapalan, 1995). Hydrological scales
are defined both by size of the area, and by locally domi-
nant processes and their representation in models. Of spe-
cial importance is the relative significance of overland flow
as compared to runoff in the channel network of the area.
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
1946 E. J. Plate: Classification of hydrological models for flood management
This is illustrated in Fig. 5. In this figure the smallest scale
is associated with the area element a (an element roughly of
1 m2 ). Surface runoff from this area element flows as over-
land or baseflow to the nearest channel. Runoff from all area
elements in sub-area Aj combine into overland flow inputs
with characteristic runoff time tC , the time of concentration
from area Aj . The total basin has n sub-areas, i.e. j =1,2,...n.
Runoff from each area Aj is routed through the channel net-
work. The process of flow routing in the channel network
has a characteristic time tf , the routing time. The ratio of
tC /tf is a suitable indicator for selection of a model scale. A
measure for tC is the rise time of the unit hydrograph for an
area, for which formulas are available, based on describing
unit hydrographs by means of a gamma function (Ihringer,
1996). A measure for tf is the flow time in the channel net-
work, expressed through length and estimated velocity of the
main channel of the network.
2.4.1 Point scale
The scale of the area element a, measured in m2 , is called the
point scale. At this scale, tf has no meaning. Processes on
this scale are mainly vertical, they are significant for flood
modeling only insofar as they determine the separation of
runoff into infiltration, surface storage (and eventual evapo-
transpiration) and overland flow. Processes on this scale are
highly non-linear and locally very variable, local soil char-
acteristics and plant cover determine local runoff, which sets
in after infitration capacity is exceeded (i.e. Bronstert, 2005).
As has been shown (i.e. Zehe et al., 2005) macro-pores in-
duce two types of switching mechanism in the conversion
of rainfall into runoff: they are activated when the infiltra-
tion capacity of the soil matrix is almost saturated and over-
land flow sets in, and they lose their effect when macro-pores
which are closed at the bottom are filled, so that only matrix
infiltration and infiltration into open macro-pores are effec-
tive. Other influences on the process of converting rainfall
into runoff are local depressions and micro-topography (de-
pression storage), which retain part of the runoff, or frost
phenomena influencing or preventing infiltration, or local ef-
fects on runoff production during snow melt. Because of
the potential variability of point scale processes for different
area elements a, processes on the point scale are considered
mostly for homogeneous areas, such as agricultural fields or
forest areas on homogeneous soils.
2.4.2 Micro-scale
Models on the point scale are building blocks for micro scale
models. The idea of subdividing a catchment into contribut-
ing areas i.e. different areas with similar infiltration poten-
tial assumes the existence of homogeneous or almost ho-
mogeneous areas, which are aggregations of area elements
a with similar infiltration characteristics. For each of these
contributing areas, the determination of the beginning of
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
sub-area Aj
area element a
tc
tf river gage
river
sub-basin
Fig. 5. Defining flow times in hydrological basins.
runoff and of an area averaged tC is important for any sub-
area Aj . RR-models
Fig.5: Definingforflow floodstimes
must consider these processes
in hydrological basins
in detail if tf /tc is small.
More generally, Plate and Zehe (2006) define micro-scales
as the scales of models which can be described by means of
the fundamental conservation laws of continuum physics, i.e.
by means of partial differential equations. Typical models of
this kind are hill slope models, which must be used if extreme
rainfall effects on surface erosion or pollutant transport have
to be considered. Extensive investigations on this scale have
been made at Karlsruhe University as summarized in Zehe et
al. (2001).
2.4.3 Meso-scale
The meso-scale is defined as that scale in which RR-models
are described by conceptual models based on system func-
tions. Basic element is one system function uj (t) for each
of the sub-areas Aj , with tcj = characteristic time of concen-
tration for area Aj . For the low mountain ranges of central
Germany this scale is associated with areas ranging from 1
to several km2 . Usually, the sub areas have small creeks,
so that tf /tc is non-zero, but their effect is included into the
concentration time. Typical for this scale are models of the
unit hydrograph type. There is no clear limit in area for
the use of unit hydrograph models; its application depends
on catchment characteristics and available data. Generally,
in flat lands unit hydrographs can be used for larger catch-
ments than in mountainous country. However, because of
lack of data hydrologists are frequently forced to use unit
hydrographs also in other large areas. On the other hand,
where data conditions are adequate (and sufficient time and
resources are available), modeling on this scale can also be
based on micro-scale or grid based models, including full hy-
draulic models of the channel network.
www.hydrol-earth-syst-sci.net/13/1939/2009/
E. J. Plate: Classification of hydrological models for flood management
The importance of meso-scale processes increases with
increasing area size. Flow in the river network increas-
ingly dominates the rainfall runoff process. In order to cap-
ture spatial differences, it is recommended today to sepa-
rate larger catchments into sub-catchments, which are con-
nected through the runoff network. Runoff from sub-areas
can be calculated by means of unit hydrographs, but other
models based on more detailed physical modeling of the rain-
fall runoff process in small catchments may also be suitable.
Models combining sub-catchment runoff with river networks
are particularly advantageous for situations in which the ge-
ological or topographic properties are very different in dif-
ferent sub-catchments, causing very different runoff forming
processes in different parts of the catchment. For example,
it becomes possible to model different reaction characteris-
tics of the rainfall runoff, depending on the time develop-
ment of local infiltration rates. This was well illustrated in
the study of Casper et al. (2003) in the Eyach-valley of the
Northern Black Forest. He could identify numerous differ-
ent sub-processes contributing to the time development of the
runoff process, with large differences in runoff formation be-
tween bogs with limited retention potential on one end of the
spectrum, and permeable sandstone on the other end.
For meso-scale processes, the spatial and/or temporal vari-
ability of the rainfall field may need to be considered. The
scale of the rainfall field is superimposed on the scale of the
catchment. Locally, it is well known that extreme rainfalls,
which cause only minor floods in a large catchment, may lo-
cally lead to very high runoff peaks, or to local flash floods.
And for large scale rainfall events, the temporal variability
of the rainfall field may completely mask the effect of the
catchment. A simple example may serve to illustrate this
point. Let the area averaged rainfall field be temporally vari-
able, as expressed through a rainfall intensity function I (t)
of duration TD , where:
Z by Ihringer et al. (1990), Ihringer (1996) and see also Plate
1
I (t) = I (t, A) dt (1)
AG
A
is the rainfall averaged over the catchment area AG . Let the
unit hydrograph for the catchment be given by a function
u(t):
u(t) = f (t, TA ) (2)
with characteristic time TA (for example, TA = rise time of
the unit hydrograph).
Assuming the runoff coefficient to be independent of
time the discharge can be determined by means of the con-
volution integral:
Zt
Q(t) = QB (t) + AG I ( ) u(t ) d (3) there is no need to model surface runoff in detail. Naturally,
0
where AG = size of the catchment, and QB (t) is the base
flow. If the characteristic time TD of the rainfall event is
www.hydrol-earth-syst-sci.net/13/1939/2009/
1947
short compared to TA , i.e. TA /TD , then I (t) can be ap-
proximated by a delta function I0 (t), and the solution of
Eq. (3) is:
Q(t) = QB (t) + I0 AG u(t) (4)
i.e. only catchment characteristics shape the discharge hy-
drograph. For the other extreme case, i.e. for TA /TD 0, the
unit hydrograph degenerates to unity, and:
Q(t) = QB (t) + AG I (t) (5)
i.e. the dynamics of the runoff process is determined exclu-
sively by the dynamics of the rainfall and by its distribution
in space, as expressed by Eq. (1). Equation (1) applies mostly
to small catchments, where an area average rainfall as calcu-
lated by Eq. (1) is appropriate. For large spatial scales, spatial
rainfall distributions as well as time distribution of area av-
eraged rainfall fields need to be considered, which requires
that the total area should be subdivided into sub-catchments
with local area averaged rainfall inputs. Such an approach
is particular important for forecasting flash floods, whereas
for design models the actual distribution of the rainfall is of
secondary importance.
As a consequence it can be concluded that meso-scale
models must be used also for cases where times tL and tF
are of the same order of magnitude. In such cases runoff for-
mation on sub-catchments is of equal importance as runoff
in the channel network. This approach should be used for
areas ranging from a few 10 s km2 to a few 100 s km2 . In
the writers experience it is particularly useful to subdivide
larger catchment into smaller catchments according to water
divides, and to describe runoff from sub areas (of a few km2
size) by means of unit hydrographs and runoff in channels
by means of 1-D calculations with St. Venant equations, or
even simpler by means of flood routing models, such as the
well known method of Kalinin-Miljukov. Our IHW-Model
et al. (1988) was developed based on this principle and has
been applied successfully for flood determinations in central
mountain regions of South and Central Germany.
2.4.4 Macro-scale
The scale of the IHW model is the scale in which runoff
generating processes associated with characteristic times tc
are of equal importance as channel flows with characteristic
time tF . With further increase of catchment size, ratio tC /tF
decreases asymptotically to 0, when dynamics of the runoff
process is fully dominated by channel flow. This is the case
of macro-scale modeling, and refers to catchment sizes rang-
ing from about 1000 to several 10 000 s km2 . On this scale,
runoff coefficients have to be determined for all sub-areas,
but retention and runoff characteristics of sub-areas can be
represented by simplified functions, such as exponential (lin-
ear reservoir) functions, or by local runoff coefficients A
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
1948 E. J. Plate: Classification of hydrological models for flood management
that are constant for each sub area A. However, for macro
scale models the importance of the hydraulic component in-
creases, which is required for converting the large discharges
of this scale into flood levels, in particular when technical
flood protection measures are planned. Such models are sub-
ject of classical hydraulics (i.e. Henderson, 1966) and not
discussed in this paper.
2.4.5 Scale transitions and uncertainties
It may seem that the best flood model would be built up by
combining point scale models into models for ever increasing
scales. Apart from the fact that this approach cannot work
for larger areas because of the sheer number of elementary
area elements to be considered, the discussion of Sect. 2.4
has made clear that this approach is not necessary because
with increase in size of the area the collective effect of the
different processes changes. For example, whereas on the
point scale local distribution of macropores may be important
(Zehe et al., 2001), they become insignificant when averaged
over large areas the law of averages stating that with increase
in the number of elements of a homogeneous ensemble the
variance of the average decreases in proportion. In the same
way, large scale features of hill slopes, such as local topogra-
phy, may become insignificant when going to a macro scale.
As a consequence, it is an interesting problem to find the
limit conditions from when on one can use models of a larger
scale i.e. to use macroscale models with lumped parameters
instead of hill slope models. In most cases this question can-
not be answered, because for larger areas the use of a model
is data driven: the availability of data dictates what type of
models can be used. Nevertheless, it is necessary that for
each location a close survey is or should be conducted, in
order to identify hydrologically significant features of every
area. For example, it should be evident that an infiltration
model (which determines groundwater recharge as well as
runoff coefficient) for sandy soils cannot be the same as an
infiltration model for karstic surfaces. The availability of in-
formation for soil and land cover has much improved over
the years through the availability of space or remote sensing
data and GIS technology, but the routinely transformation of
space information into hydrological parameters has yet to be
developed. At this time, the translation of satellite informa-
tion into soil parameters etc. depends on ground truth obser-
vations. This adds to the uncertainty of models, and is part of
the epistemic uncertainty with which the modeler has to live.
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
3 Comparison of flood models for planning and
forecasting
3.1 Advantages and disadvantages of flood models for
planning
The principal statistical tool for planning flood protection
works is extreme value analysis. Extremes of runoff are
determined by two independent methods (which ideally are
complementing one another). The first and more traditional
method uses directly extreme values of time series for dis-
charges (or water levels) for determination of design floods
typical is a flood with a recurrence interval of once every 100
years i.e. the 100 year flood. There are a number of disad-
vantages to this approach. It relies on measured time series
of water discharges (or water levels) at a point, i.e. it obvi-
ously requires presence of a gage on the river near the spot
for which one wants to determine the design flood. For sta-
tistical reasons it needs observations of long time series for
a reasonable fit to a generally unknown probability dis-
tribution of extremes. The most important advantage of this
method is that it avoids uncertainties of the generating pro-
cess of the extremes. There exists a vast literature on extreme
value analysis, which shall not be discussed here.
The second method uses RR-models. They also depend on
statistical inputs, this time of rainfall fields. Uncertainty of
runoff prediction from RR-models primarily stems from pre-
diction of the extreme rainfall event for the catchment, i.e.
the inherent uncertainties of extents and intensities of rain-
fall fields. Additional uncertainties are caused by the time
variability of soil moisture and other dynamic catchment pa-
rameters needed to convert rainfall into runoff. Advantages
of RR-models are obvious. Rainfall inputs are less dependent
on local conditions, and thus rainfall statistics can be general-
ized for large areas. Furthermore well calibrated RR-models
can be used for flood prediction not only peak values at
every point in a catchment.
Because different extreme value distributions applied to
the same data set may yield very different extreme flood peak
values, it is actually not sufficient if only the recurrence in-
terval is specified. For completeness of specifications, also
the method of determination of the extreme values should
be given, as for example in the recommendations of the US
Interagency Advisory Committee on Water Data (IACWD,
1982). However, the actual recurrence interval for design
floods is never accurately known because of numerous po-
tential errors due to model complexity (model error), incom-
plete information on parameters (parameter error) or insuffi-
cient or inaccurate data (data and sampling errors). The true
recurrence interval of an observed flood peak can never really
be ascertained. An error range of 15%=30% is not uncom-
mon, and this may mean for example when determining
the 100 year flood to design not for the 100 year flood, but
for floods of recurrence interval of 150 or 50 year. Already
small changes in the flood level may cause large changes in
www.hydrol-earth-syst-sci.net/13/1939/2009/
E. J. Plate: Classification of hydrological models for flood management 1949

calculated probability. This is illustrated in Fig. 6 where ra- 1600 1600

tios x100 /x20 (i.e. of the 100-year to the 20 year flood), and
1500 1500
ratios x1000 /x100 (i.e. of the 1000 year flood to the 100 year
flood) are shown as function of coefficient of variation CVx 1400 1400

x1000/x100
x100/x20
for the two parameter gamma distribution. (This distribu-
tion has been found give best fits to long data series of larger 1300 1300

basins in South Germany, LfU, 1999). For two parameter 1200 1200
distributions, such a presentation is unique, for three param-
eter distributions one obtains a family of curves with skew 1100 1100
coefficient CSx as third parameter.
1000 1000
For two parameter gamma distributions CSx and CVx are 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 1,1
related as CSx =2CVx , where the magnitude of CSx is re- Variationskoeffizient CV
stricted by the fact that already for CSx =2 gamma distribu-Fig.6: Ratios Q1000/Q100 and Q100/Q20 (multiplied by factor 1000) as function of coefficient of
tions reduce to the exponential. Thus, if one assumes an av- Fig. 6.variability
Ratios Q CV1000
/Q100 and Q100 /Q20 (multiplied by factor
x of the two-parameter gamma-distribution
erage skew factor of 1, corresponding a CVx =0.5, one finds 1000) as function of coefficient of variability CVx of the two-
both curves to show differences of 30% between the 1000 parameter gamma-distribution.
and 100 year flood, or between the 20 year flood and the 100
year flood, respectively, corresponding to a tenfold resp. five-
fold range of return periods. This uncertainty has epistemic a more frequent, or a less frequent event. The case of flood
and natural causes. Epistemic uncertainty includes both data forecasting is different, after a flood one knows for certain
and model uncertainty, whereas natural variability is due to that the forecast has been accurate, or accurate enough. On
the complexity of the natural processes and catchment char- the other hand, a forecast model does not have to correctly
acteristics leading to runoff variability. These uncertainties model the physical basis of the rainfall-runoff process, so that
add up to a wide range of potential exceedance probabilities, any method which is reasonable and yields acceptable results
as was well illustrated in a number of examples in a recent may be used, for example those from regression analysis, or
paper by Merz and Thieken (2005). from artificial neural networks which have been trained on
past records. The overriding concern is to include the error
It is very difficult to overcome these uncertainties. Some
band in the discussion of the results, i.e. to use the probability
improvement is found by regionalization based on many
distribution of potential outcomes as basis for a purpose ori-
different runoff gages in a region. Regionalization was
ented forecast, for example take the ensemble average as best
also used by Ihringer (LfU, 2005) who developed a region-
estimate, and give error bands based on exceedance probabil-
alization model, which permits to estimate the 100 year
ities.
flood peaks for every point in the German State of Baden-
Wurttemberg. But it must be realized that in the end the de- What is meant by a good forecast has to be specified not
cision for flood protection measures is a political decision, only intuitively (by how the warned PAR (= people at risk)
which leads to a politically acceptable recurrence interval, feel about the forecast) but objectively, by means of an ob-
typically based on large historical floods. For example, the jective criterion, which is derived from past events of varying
design flood for the Upper Rhine (Rhine between Basel in magnitude.of varying magnitude. Many different statistical
Switzerland and Mannheim at the confluence of Rhine and quantities have been suggested and used, (for some recent
Neckar) has approximately a T =200 years, but it is based on uses see Bravo et al., 2009, or Wu et al., 2005) which do not
the extreme value observed in 1882, (shifted in time to ac- really meet this requirement. Obviously, a criterion based on
count for Rhine corrections made after 1882, so that peaks the average performance of a flood hydrograph, such as the
of Neckar and Rhine floods coincide), plus freeboard. This criterion by Nash and Sutcliffe (1970) is not sufficient. The
approach may appear rather simple in terms of modern hy- criterion must be geared to weigh the course of the hydro-
graph of the future, as observed from a known point of the 22
drology, but it had the advantage of being plausible and po-
litically acceptable. hydrograph of the past. A possible criterion could be based
on the following considerations.
3.2 Advantages and disadvantages of forecast models It is evident that for short times (short in relation to a crit-
ical time, which depends on the size of the catchment or
The major difference between forecasting and planning mod- basin) the requirements for complexity of forecast models
els is accuracy. Flood forecast models require higher accu- are not very high because the discharge cannot have dis-
racy than planning models. Flood forecast models require continuities, for physical reasons. If there is no other infor-
that an exact peak value is forecast, in contrast to results from mation available, the best forecast for the near future is to
planning models, as has been discussed in 3.1. It will never forecast the value at time t0 +TF as being equal to the value
be possible to accurately identify the recurrence interval of an at t0 . The performance of a forecast, expressed through a
extreme event that has actually happened. It could have been forecasted value hf (t+TF ) after time TF should be judged

www.hydrol-earth-syst-sci.net/13/1939/2009/ Hydrol. Earth Syst. Sci., 13, 19391951, 2009

1950 E. J. Plate: Classification of hydrological models for flood management
relative to this value. This implies that the deviation of the
forecasted value hf (t+TF )h0 (t) from the present value
should be large relative to the deviation of the actual value
ha (t+TF ) from the forecasted value hf (t+TF ), i.e. from
ha (t+TF )hf (t+TF ). In terms of quadratic deviations (to
eliminate the need for sign corrections) this condition is ex-
pressed by an index (Kitanides and Bras, 1980)1 :
P 2
hf (t + TF ) ha (t + TF )
n
IF (TF ) = 1 P 2 (6)
hf (t + TF ) h0 (t)
n
for which during calibration the sums have to be taken at each
time t=i1t, i=1, 2, ...n over the whole forecast interval
TF =n1t of the event, where 1t is the time increment. A
positive value close to 1 of IF (TF ) indicates good, a small or
even negative value poor performance, i.e. if IF (TF ) is close
to 0 or even negative, the performance of the forecast is not
better than taking the value of today (at time t) as forecast for
the value at time t+TF . It is concluded that forecast models
are even more dependent on a good data base than planning
models. Only many comparisons of actual with forecasted
data can establish confidence in a forecast model.
4 Conclusions
This paper makes a case for considering models for flood
forecasting and models for planning of flood protection struc-
tures as important but quite different tools for managing flood
risks. Flood risk management is seen as a comprehensive
approach for handling the consequences of extreme flood
events so that they do not lead to flood disasters. The ob-
jective of development or adaptation of models is their in-
tended application. This requires that we distinguish models
by scale and by topographic context: models for large basins
in topographically flat country require different approaches
than, for example, models for flash floods in mountain areas.
As prerequisite of model choice a thorough understanding
of local topography and climate processes is essential. But
models should not only reflect local scales and local terrain
features and geology, but they also should be determined by
the intended application. By pointing out the different con-
ditions and requirements for the use of flood management
models, this paper is an attempt to give guidance to persons
involved in flood management.
Differences in the types of models for planning versus
models for flood forecasting are stressed, although the pa-
per is descriptive in nature. Analytical concepts for model
structures and calculations cannot be covered, and for details
references are given which reflect mostly the experience of
1 Note that this is similar to the criterion of Nash and Sut-
cliffe (1970), except that height h replaces discharge Q, and the
initial value h0 (t) at time t replaces the average value of the com-
plete hydrograph.
Hydrol. Earth Syst. Sci., 13, 19391951, 2009
the author and his team at the University of Karlsruhe. Hy-
drologists today must be aware that their work can serve two
very different purposes: one is to better understand nature,
and the other is to provide analytical tools for helping water
managers and design engineers to better handle their design
and management problems. A better understanding of na-
ture is obtained by observing, measuring, and modeling of
hydrological processes at the smallest scales and by extend-
ing the resulting models to larger ones by integration over
more and more small scale area elements. Engineers, on the
other hand, need models which reproduce only those model
components which are relevant to quantify the runoff pro-
cess with sufficient accuracy for their purposes. In looking
at planning models and forecast models for floods, the two
different aspects of modeling become particularly evident.
Acknowledgements. The thoughtful and extensive discussions of
an earlier draft of the paper by E. Todini and G. Bloschl are very
much appreciated.
Edited by: H. H. G. Savenije
References
Betamio de Almeida, A. and Viseu, T. (Eds.): Dams and Safety
Management at Downstream Valleys, Balkema, Rotterdam,
247 pp., 1997.
Beven, K. and Freer, J.: Equifinality, data assimilation, and estima-
tion in mechanistic modeling of complex environmental systems
using the GLUE methodology, J. Hydrol., 249, 1129, 2001.
Bloschl, G. and Sivapalan, M.: Scale issues in hydrological model-
ing: a review, Hydrol. Process., 9, 251290, 1995.
Bloschl, G., Grayson, R. B., and Sivapalan, M.: On the represen-
tative elementary area (REA) concept and its utility in rainfall
runoff modelling, Hydrol. Process., 9, 313330, 1995.
Bravo, J. M., Paz,A. R., Collischonn, W., Uvo, C. B., Pedrollo,
O. C., and Chou, S. C.: Incorporating forecasts of rainfall in two
hydrological models used for medium range streamflow forecast-
ing, J. Hydrological Engineering, ASCE, 14, 435445, 2009
Bronstert, A. (Eds.): Abflussbildung: Prozessbeschreibung und
Fallbeispiele, Forum fur Hydrologie und Wasserwirtschaft, DWA
Deutsche Vereinigung fur Wasserwirtschaft, Abwasser und Ab-
fall Fachgemeinschaft Hydrologische Wissenschaften, 137 pp.,
2005.
Casper, M., Volkmann, H. N., Waldenmeyer, G., and Plate, E. J.:
The separation of low pathways in a sandstone catchment of the
Northern Black Forest using DOC and a nested approach, Phys.
Chem. Earth, 28, 269275, 2003.
Crowley II, T. E. and He, C.: Distributed parameter large basin
runoff model I: Model development ASCE, J. Hydrol. Eng., 10,
173181, 2005.
Dooge, J. C. I.: Scale problems in hydrology, in: Reflections on Hy-
drology, edited by: N. Buras, N., American Geophysical Union,
Washington, DC, USA, 85135, 1997.
De Roo, A. P. J., Wesseling, C. G., and van Doysen, W. P. A.: Phys-
ically based river basin modelling within a GIS platform: the
model LISFLOOD, Hydrol. Process., 14, 19811992, 2000.
www.hydrol-earth-syst-sci.net/13/1939/2009/
E. J. Plate: Classification of hydrological models for flood management
EU: European Union Directive 2000/60/EC of the European Par-
liament and of the Council, http://eurlex.europa.eu/LexUriServ/
LexUriServ.do?uri=CONSLEG:2000L0060:20011216:EN:PDF,
2000.
EU: European Union Directive 2007/60/EC of the European Par-
liament and of the Council, http://eurlex.europa.eu/LexUriServ/
LexUriServ.do?uri=OJ:L:2007:288:0027:0034:EN:PDF, 2007.
Garbrecht, J. and Martz, L. W.: The assignment of drainage direc-
tion over flat surfaces in raster digital elevation models, J. Hy-
drol., 193, 204213, 1997.
Henderson, F.M.: Open channel flow, Macmillan, New York,
522 pp., 1966.
IACWD (Interagency Advisory Committee on Water Data): Bul-
letin 17-B Guidelines for determining flood flow frequency:
United States Geological Survey, Washington, DC, 1982.
Ihringer, J.: Hochwasser aus landlichen und stadtischen Gebieten,
Z. Geol. Wissenschaft., 14, 523530, 1996.
Ihringer, J., Beelitz, W. D., Gunther, Ch., and Kron, W.: Soft-
warepaket fur Hydrologie und Wasserwirtschaft, Anwender-
handbuch, Vol. 1: Hochwasseranalyse, Universitat Karlsruhe, In-
stitut fur Hydrologie und Wasserwirtschaft, 1990.
Kitanides, P. and Bras, R.: Real time area forecasting with a concep-
tual hydrological model. 1. Analysis of uncertainty, Water Re-
sour. Res., 16, 10251033, 1980.
Krzysztofowicz, R.: The case for probabilistic forecasting in hy-
drology, J.Hydrol., 249, 29, 2001.
LfU: Hochwasserabflusswahrscheinlichkeiten in Baden-
Wurttemberg, Landesanstalt fur Umweltschutz, Baden-
Wurttemberg, Oberirdische Gewasser, Gewasserokologie UN/ISDR: United Nations Inter-Agency Secretariat of the Interna-
No. 54, 45 pp., 1999.
LfU: Leitfaden Festlegung des Bemessungshochwassers bei An-
lagen des technischen Hochwasserschutzes Landesanstalt fur
Umweltschutz, Baden-Wurttemberg, 91 pp., 2005.
Loucks, D. P. and van Beek, E.: Water Resources System Planning,
Studies and Reports in Hydrology, UNESCO, Paris, 2005.
Merz, B. and Gocht, M.: Karten fur die Hochwasservorsorge und
das Risikomanagement auf der lokalen Skale, Hydrol. Wasserbe-
wirts., 47, 186194, 2003.
Merz, B. and Thieken, A.: Separating natural and epistemic un-
certainty in flood frequency analysis, J. Hydrol., 309, 114139,
2005.
Nash, I. E. and Sutcliffe, J. V.: River flow forecasting through con-
ceptual models, 1., A discussion of principles, J. Hydrol., 10,
282290, 1970.
www.hydrol-earth-syst-sci.net/13/1939/2009/
1951
Plate, E. J., Ihringer, J., and Lutz, W. L.: Operational models for
flood forecasting, J. Hydrol., 100, 489506, 1988.
Plate, E. J.: Skalen in der Hydrologie, Zur Definition von Begrif-
fen, Regionalisierung hydrologischer Parameter, Beitragssamm-
lung Deutsche Forschungsgemeinschaft, Bonn, Germany, 3344,
1992.
Plate, E. J.: Flood risk and flood risk management, J. Hydrol., 267,
211, 2000.
Plate, E. J.: Early warning and flood forecasting for large rivers with
the lower Mekong as example, Journal of Hydro-Environment
Research, 1, 8094. 2007.
Plate, E. J. and Zehe, E. (Eds.): Wasser- und Stofftransport aus
kleinen landlichen Einzugsgebieten, 365 pp., Schweizerbart,
Stuttgart, Germany, 2008.
Refsgaard, J. C. and Storm, B.: MIKE SHE, in: Computer Mod-
els of Watershed Hydrology, edited by: Singh, V. P., Water Re-
sources Publications, 809846, 1995.
Schumann, A. H., Antl, M., and Geyer, J.: Probabilistische Ab-
schatzung der Unsicherheit von Aussagen zu hydrologischen
Auswirkungen globaler Klimaveranderungen, Hydrologie und
Wasserbewirtschaftung, 45(4), 154165, 2001.
Singh, V. J. and Woolhiser, D. A. : Mathematical modeling of wa-
tershed hydrology, J. Hydrological Engineering, ASCE, 7, 270
291, 2000.
Todini, E.: The ARNO rainfall runoff model, J. Hydrol., 175,
339382, 1996.
Todini, E.: Role and treatment of uncertainty in real time flood fore-
casting, Hydrol. Process., 18, 27432746, 2004.
tional Strategy for Disaster Reduction (ISDR), Living with risk:
a global review of disaster reduction initiatives, United Nations
Publications, Geneva, 2004.
Wu, J. S., Hun, J., Annambhotia, S. and Bryant, S.: Artificial neu-
ral networks for forecasting watershed runoff and stream flow, J.
Hydrological Engineering, ASCE, 10, 216222, 2005.
Zehe, E., Maurer, T., Ihringer, J., and Plate, E. J.: Modelling wa-
ter flow and mass transport in a Loess catchment, Phys. Chem.
Earth, Part B, 26, 487507, 2001.
Zehe, E., Becker, R., Bardossy, A., and Plate, E. J.: Uncertainty of
simulated catchment runoff response in the presence of thresh-
old processes: Role of initial soil moisture and precipitation, J.
Hydrol., 315, 183202, 2005.
Hydrol. Earth Syst. Sci., 13, 19391951, 2009